What this section is for
An agent that remembers your preferences is a different tool from one that starts cold every session. Hermes Agent keeps that state in a handful of places on disk - a memory file, a skills tree, cron entries, a session database - and all of them grow. This section measures that growth instead of describing it, because the numbers are the part that decides whether you have a problem.
Why size matters more than the feature list
Memory is not unbounded storage. The main memory file has a character budget, and when it is full the agent does not stop working - it starts consolidating, dropping and merging entries so the next session still fits. That is a real behaviour with a real cost: the entry that gets dropped is often the one you never re-read. The interesting question is not whether it has memory, but what gets lost first, and how you would notice.
Pages in this section
- Three stores, one budget - the memory file, the skills tree and the session database, with the character counts and disk numbers we measured on this machine.
What we will not publish here
The contents of our own memory. It holds operational details that belong to other projects, and quoting it would be a worse article for a reader and a bad trade for whoever it describes. We publish sizes, structure and behaviour - not the text.
How to check your own install
wc -c MEMORY.md USER.md- how close you are to the budget, in characters.ls -1 "$LOCALAPPDATA/hermes"- the top-level layout the installer created (Windows, git-bash).du -sh "$LOCALAPPDATA/hermes/state.db"- how much disk the session database has taken.
If the memory file is above roughly four fifths of its budget, expect the agent to start consolidating, and read what it removed before you assume it kept everything.
